Design And Implementation Of A MD5Parallel Crack Algorithm Based On CUDA

With the continuous development of computer technology, the computer has beengradually integrated into people’s work and daily life. Due to more and more computercrimes cases, a large number of computer-related criminal investigations and evidencegathering are needed to obtain electronic evidence in computer. In the process ofinvestigation and evidence collection, investigators often are required to decipher theencrypted data or system security password, which requires a lot of password crackingwork. As a commonly used encryption algorithm, MD5algorithm isused in many areasof computer security. Because the MD5algorithm is a high-strength encryptionalgorithm, the crack in the past which just used the computing power of the CPU, wasinefficient.With the computer graphics processor GPU entering into the field ofgeneral-purpose computing, a SMP cluster which equipped with a number of blocksGPU graphics computing module came out in the field of high-performance computers.In order to improve the efficiency of the MD5algorithm crack,we naturally considerwhether the MD5high-speed parallel crack system can be realized on the GPU-basedSMP cluster to take full advantage of SMP clusters and GPU computing power toachieve a rapid crack of the MD5algorithm.This article analysizes in detail the high-performance parallel computingtechnology, the new Fermi architecture GPU and CUDA Compute Unified DeviceArchitecture, then it futhur analysizes the advantages of the GPU-based SMP cluster inhigh-performance computing. Through a careful study of the new GPU hardwarecharacteristics which adopts the Fermi architecture, we summarizes the methods andthe advantages of using CUDA programming to determine the MPI-CUDA GPU-basedSMP cluster system using double-hybrid parallel programming model developed theMD5high-speed parallel crack system. To cater for the actual demand,we study indetail the MD5algorithm and its crack,then analysize the parallelizability of thecracking algorithm,elaborate how to dismantle the password cracking task, and design the program flow of the crack system.In the implementation part,this article specifically described how to takeadvantage of the MPI-CUDA parallel hybrid programming model to programme on theGPU-based SMP cluster system. Finally, through the experiment we test the system,found the deficiencies in the system, then optimize the cracking system, andsummarize and analyze the advantages and disadvantages of the MPI-CUDAprogramming model and its program optimization method. The final validation resultsis that the efficiency of using the MD5algorithm to crack the system is35times thantraditional CPU-based crack algorithm, and up to95times after optimization.